Ok .. to clear things up .. I am _not_ interested in better libraries or algorithms for that problem.
I used bigint because I was not sure whether one of the Collatz chains would produce an intermediate number exceeding the int scope (probably my take on pessimation). Only after reading on the projecteuler forum that everyones solution was running in 1-5sec and mine taking 10min - while using the same algorithm as many others - made me suspicious.
So I removed "use bigint" .. lo, and behold! It runs .. fast. Googling "perl bigint performance" didn't turn up anything useful, that's why I asked here.
PS. I used pastebin because I wasn't sure about the "post code" conventions ;-)